Real-Time Position Detecting of Large-Area CNT-based Tactile Sensors based on Artificial Intelligence

Abstract

For medical device and artificial skin applications, etc., large-area tactile sensors have attracted strong interest as a key technology. However, only complex and expensive manufacturing methods such as fine pattern alignment technology have been considered. To replace the existing smart sensor, which has to go through a complicated process, a new approach including a simple piezoresistive patch based on artificial intelligence has been suggested. Specifically, a 16-electrode terminal was connected to the edge of a polydimethylsiloxane pad where multi-walled carbon nanotube sheets are well dispersed, and a voltage input to the center of the specimen. The collected data was calculated using a voltage divider circuit to collect the voltage data. 54 random positions were marked on the pad. 4 positions were configured as the validation data set and 50 positions as the training data set. We examined whether it was possible to determine points in untrained positions using a deep neural network (DNN) and 12 different machine learning (ML) algorithms. The result of a deep neural network for untrained point location identification was MSE: 0.00026, R2: 0.991158, and the result of Random Forest, an ensemble model among ML algorithms, was MSE: 0.00845, R2: 0.971239. Real-time position detection is possible using smart sensors created by combining simple bulk materials and artificial intelligence models from research results.